Babar%Ali%(NHSC)%herschel.esac.esa.int/...BAli_MapMaking2013_DPWS.pdfAli-PACS-preProc.ppt Author:...

page 1 MapMaking Workshop 2013

PACS pre-‐processing MADmap.

Preprocesing PACS photometer 2melines for map-‐making with MADmap

Babar Ali (NHSC)



MADmap in one slide

•  Mapping code wri>en by the Berkeley CMB group to remove 1/f noise from bolometers. h>p://newscenter.lbl.gov/feature-‐stories/2010/02/03/madmap/

•  MADmap was ported to Java for use in HIPE. •  MADmap offers the so-‐called opPmal map-‐making to convert Pme ordered readouts to a final map. –  Uses maximum likelihood (given a noise/probability model) to determine the opPmal sky value.

•  Details in the next talk.



The need for preprocessing.

•  MADmap assumes that bolometer Pme-‐lines are relaPvely calibrated. – No pixel-‐to-‐pixel variaPons

•  MADmap assumes that 1/f noise is uncorrelated amongst pixels.

•  PACS Level 1 Pmelines contain all of the above.



Documenta2on Reference

•  PACS data reducPon guide, chapter 9 •  NHSC Tutorials:

– PACS-‐101: IntroducPon to PACS tutorials – PACS-‐103: Accessing & Storing PACS data – PACS-‐104: Using iPipe scripts – PACS-‐201: Level 0 to 1 processing of PACS photometer data

– PACS-‐401: MADmap map-‐making. h>ps://nhscsci.ipac.caltech.edu/sc/index.php/Pacs/DataProcessing



Preprocessing Steps

I.  Remove pixel-‐to-‐pixel electronic offsets (addiPve). II.  Remove correlated signal driZ (addiPve). III.  IdenPfy and fix module/row/pixel anomalies.



I: Pixel-to-pixel offsets Additive correction



Electronic Offsets

•  “StarPng point” of the raw readout value for each pixel is different and addiPvely offset.

•  Usually, the offset pa>ern is the only visible structure in raw PACS readouts

•  Correct by using calibraPon file, or median of the enPre Pmeline for the pixel.

!

Raw Corrected



II: Correlated Signal Drift Additive correction



Proper2es of Signal DriH

•  The median value of the en2re array shows a strong correla2on with 2me.

•  The trend is usually more dominant than astrophysical signal.

Median vs. Time When present, this is the strongest trend in PACS cubes



Proper2es (cont.)

•  No model has been idenPfied (so far) to explain said behavior. – Caused by temperature varia0ons? But, no thermometers available to confirm or reject.

– Caused by electronics? But, not correlated with suspected Housekeeping values tried.

•  The trend is strongest at the start of PACS OD and driZs to negligible values by the end of OD



Proper2es (cont.)

•  The trend is also different from module to module.

(next slide)



Examples of signal driH color coded by individual PACS 16x16 pixel modules

“Usual”

Strong module-to-module differences

Rare. Ascending

“Hook”



Proper2es (cont.)

•  There is now evidence that individual pixels or groups of pixels have 2nd-‐order driZs on Pme scales larger than used for 1/f correcPon.

•  Will require separate pre-‐processing •  Or, increased noise filters lengths



Correc2ng Signal DriH

•  First step is to define the baseline.



Es2ma2ng DriH Baseline

Use MEDIAN to pick up the correlation.

Astronomical source



Es2ma2ng DriH Baseline

Divide MEDIAN values in bins.

Size of bin.



Es2ma2ng DriH baseline

Take MINIMUM of bins to reject sources. Fit MINIMUM values.



DriH Correc2on Modules

•  photGlobalDri9Correc0on – 1st generaPon. – Same correcPon for enPre FOV – Applies a single polynomial fit to baselines.

•  photBaselineDri9Corr – 2nd generaPon. – Correct module-‐by-‐module – Segment baseline for spline-‐like ficng instead of a single polynomial



DriH Correc2on Modules

•  photSignalDri9Corr – 3rd generaPon – Development version. – Fits the local neighborhood only

•  for be>er esPmate of “hooks” and similar structures.

– Adds an iteraPve sigma-‐clipper to remove deviant points.

– Adds an “outer iteraPon”



Width is the number of +/- points used to calculated the local baseline

Larger widths = smoother baseline



“Outer Itera2on”

Determine baseline

Project Naivemap

Naivemap Signal Timelines

Subtract Signal Timeline from Original L1 Timelines

Original L1 Timelines

Subtract baseline

Baseline Timelines Per pixel

Smooth Baseline Timelines

Subtract Baseline Timelines from Original L1 Timelines

Project MADmap



3rd gen. results

Good, but not fully tested.



Proper correc2on is crucial

Bad

Good



DriH Correc2on

•  Preprocessing remains very interacPve. – Less so with 3rd gen code.

•  Crucial to detect and mask remove bad data and avoid pidalls.



III. Masking bad pixels



Calibra2on block echoes

No correction (yet) in preprocessing.

Unstable frames: Usually at the beginning Likely due to calibration pre-amble.

Fix: mask range of readouts as “bad”



Clean up

Artifacts on projected maps are usually issues in readouts (next slide).



Clean up (cont.)

Example: Single pixel that went berserk.

Not a single readout, but 1000s of readouts



Clean up (cont.)

Example: An entire row



Fixes

•  There is no PACS pipeline module for automaPcally detecPng these signal anomalies.

•  IdenPfy by examining the final map, then reproject to Pmelines to idenPfy pixels and readout range.

•  Masking the affected readouts as “bad glitches” remains the only opPon.



IV. Post-processing Correct the final map for point source

artifacts



Point-‐source ar2facts

Original map

Corrected map

Artifact map Naïve

map



How does it work?

•  Reference: – Lorenzo et al. in prep (available on request) – Developed for HiGal data processing.

•  Uses the opPmal map as the starPng point for sky emission. •  Subtract sky from original Pme lines, which leaves only noise

+arPfacts in the Pme-‐line. •  High-‐pass filter the noise+arPfacts Pme-‐lines. This removes

noise and leaves on arPfact. •  In pracPce, several iteraPons are necessary.

Date post:	21-Sep-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Babar%Ali%(NHSC)%herschel.esac.esa.int/...BAli_MapMaking2013_DPWS.pdfAli-PACS-preProc.ppt Author:...

Documents